Solving Heart Rate Data Inaccuracies


Heart rate monitors are useful for athletes, researchers, and clinicians as a method of assessing fitness and cardiovascular health. Heart rate is a basic measure that clinicians use to monitor patients and typically do so with electrocardiogram (ECG). ECG is not as easy to measure upon movement and often requires larger, non-portable measurement devices. For athletes who seek to measure heart rate during exercise, these heart rate monitors are found in the form of a watch or chest strap. Despite its wide-use and high level of reporting, user activity can affect heart rate monitor accuracy. For LED heart rate monitors, where a light is emitted onto the user’s skin as opposed to electrodes that rest on the surface of the skin in ECG, position of the device on the body (Figure 1) and user activity can potentially affect the heart rate data. The issue of erroneous data is especially prevalent for multisport athletes who engage in activities that require different body positions and movements.

Obtaining consistent heart rate data collection and analysis is essential for being able to compare data across activities. For example, an athlete may have a reported average heart rate over a cycling activity as 110bpm and 140bpm during a running activity. Is this difference is due to different intensities for each activity or is it due to inconsistencies in data collection due to different noise factors dependent on activity?

Being able to identify characteristic noise frequencies based on user activity and monitor placement can greatly benefit athletes. This will allow users to either change the placement of the device based on their activity or, more favorably, input the activity to the device prior to data collection, allowing the device to prepare different filtering techniques based on expected noise for that particular activity. Applying these filters to the device would enhance heart rate monitor design and better ensure accurate data.


Before solving this problem, a basic understanding of heart rate monitors is necessary. LED heart rate monitors emit a green light to the user’s skin. Some of this light is absorbed, and some is reflected back into a receiver within the device. The absorption can be modeled using the Beer-Lambert law, A =  ε𝓁c, where absorbance (A) is equal to absorptivity (ε), beam length (𝓁), and concentration of absorbing species  (c) [4].

The intensity of the light, based on absorbance, is evaluated by a photodiode in the receiver and is transformed into a photoplethysmogram (PPG) signal by the device’s processor. This PPG signal contains cardiac, respiratory, and motion data, as well as noise. Filters act to remove non-cardiac components from the PPG signal [1] – this is where heart rate data error can occur if not enough noise is removed (Figure 2). Since motion is a noise component in the PPG signal, heart rate data can be influenced by monitor position and user activity [2].

To assess potential discrepancies in heart rate data based on monitor position and user activity, two heart rate monitors can be worn in different locations during various activities (Figure 1). In data collection, it is important to know heart rate data for each of these monitor placements over time at rest and during different activities. It is also important to use a ECG monitor as a reference (i.e a verified, widely-used chest strap) to know what heart rate data should be.

Assumptions made in solving this issue is that light intensity emitted from the heart rate monitor device is consistent throughout an activity, all activities, and between devices. Another assumption is that user intensity for each activity is the same such that heart rate increases are due to monitoring differences rather than increased workout intensity. PPG signal will also be assumed to only have components of cardiac (i.e. heart rate) signals and non-cardiac noise signals. These assumptions allow for heart rate output due to device position and activity to be determined.


Solving the problem of heart rate monitor outputs varying based on user activity and device position first requires an understanding of which device placements respond to different activities. This data is collected from users wearing monitors in different positions and engaging in different activities. It has been determined that wrist-based measurements respond better to walking and running activities than forearm measurements while forearm measurements respond better to cycling activities than wrist measurements (Figure 3). Both respond well to rest [2].

Next, an analysis of the PPG signal generated for each device placement and each activity can be completed. This allows for the determination of frequencies that are characteristic of noise in each of these activities. For example, since cycling creates more noise in wrist measurements than forearm measurements, the PPG signals of each of these measurements can be compared to determine what is present in the wrist signal that is not present in the forearm signal. This discrepancy, say frequency A, is noise generated in wrist measurements due to cycling motion.

Performing this analysis on a large data set of heart rate measurements over different device placements and user activities allows for the identification of noise frequencies characteristic of certain user activities. In designing a heart rate monitor device, these frequencies can be programmed to be filtered out based on the activity being completed. For example, a user can select “cycling” on their wrist based heart rate monitor prior to beginning exercise and the device will then apply the filter to remove frequency A noise from the data, producing heart rate data just as accurate as a forearm monitor. Designing devices such that noise can be correctly identified and removed will allow multisport athletes to gain reliable heart rate data regardless of activity and without having to move their device on their arm.

This solution is reasonable, however determining exact frequencies characteristic of certain activities would require large data sampling since there is much individual variation in heart rate. Limitations stem from assumptions mentioned above, as user intensity is often not constant across activities and emitted light intensity may change throughout an activity as the device moves closer and farther from the user’s skin due to motion.


[1]       P. R. MacDonald and C. J. Kulach, “Heart Rate Monitoring with Time Varying Linear Filtering,” US 9801587 B2, 2017.

[2]       J. Parak and I. Korhonen, “Evaluation of wearable consumer heart rate monitors based on photopletysmography,” 2014 36th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. EMBC 2014, pp. 3670–3673, 2014, doi: 10.1109/EMBC.2014.6944419.

[3]       J. L. Cheng, J. R. Jeng, and Z. W. Chiang, “Heart rate measurement in the presence of noises,” 2006 Pervasive Heal. Conf. Work. PervasiveHealth, 2006, doi: 10.1109/PCTHEALTH.2006.361651.

[4]       D. F. Swinehart, “The Beer-Lambert law,” J. Chem. Educ., vol. 39, no. 7, pp. 333–335, 1962, doi: 10.1021/ed039p333.



How Garmin Watch Heart Rate Monitors Work

Using a GPS watch has become the norm in distance running. These watches provide users with information regarding distance traveled, pace, and even maps of the route taken. Newer watches also include heart rate monitors, providing users with greater information about their fitness. The popular watch brand, Garmin, has a patented heart rate monitor [1] used in their watches, seen in Figure 1 below. 

Figure 1. Back of Garmin watch with heart rate monitor device (labeled “610”) [1].

The heart rate monitor in Garmin watches monitors cardiac signals via the user’s wrist. The main claims of this invention are as follows:

  • The device consists of an emitter, receiver, inertial sensor, and time-variant sensor. The processor determines frequency associated with the motion signal, transforms the signal from PPG into the frequency domain, identifies the cardiac component of the PPG signal, configures a time-variant filter, and calculates the time between cardiac component cycles.
  • The device emits a light signal and receives an input of the light’s reflection, which eventually allows for the isolation of the cardiac component of signal.
  • The cardiac component of signal allows for heart rate to be determined.
  • The time between successive cycles gives insight into heart rate variability, stress, recovery time, VO2 max, and/or sleep quality.
  • The device contains an interface that displays determined information to the user.

This device would be of interest to any Garmin watch user, especially those interested in heart rate during exercise. This watch, primarily used by runners, tells the user their heart rate and therefore how fast their heart is pumping blood through the body at any given time during exercise. This gives insight into the user’s fitness and exertion levels and ensures the user is in desired heart rate zones while training. Knowing how heart rate changes personally affect the user can also give insight into dehydration, stress, and needed recovery. Using this device over an extended period of time allows for users to see improvements in heart rate due to exercise.

How Does it Work?

The heart rate monitor in Garmin watches directs light from a light-emitting diode (LED) to the skin of the user. The reflection of the light is received by a photodiode, which sends a light intensity signal to the processor. The processor generates a photoplethysmogram (PPG) signal – containing cardiac, motion (determined by an inertial sensor, which senses movement of the device), and respiratory components – based on the intensity of the reflected light.

To isolate the cardiac component of the PPG signal, time-variant filters are used to remove non-cardiac components. The PPG signal can initially be filtered with a bandpass filter that only passes signals within the range of possible cardiac component frequencies. This bandwidth can be adjusted by the processor to account for lesser or greater expected cardiac frequencies based on changes in the environment. For example, if the user begins running, the processor senses rapid motion change and the bandwidth will increase since heart rate is expected to rise.

To determine which other signals to remove within the passband, the processor first identifies one or more frequencies associated with the motion signal via the inertial sensor. The processor then transforms the PPG signal into the frequency domain. Comparing the identified motion signal frequencies with the transformed PPG signal allows for the cardiac component of the signal to be determined within the frequency domain. Then, based on the identified cardiac component, the processor is able to determine filter coefficients for the cardiac component which are configured into the time-variant filter. When the PPG signal is transformed back into the time domain and filtered through this time-variant filter, the motion component is removed from the PPG signal. This results in a time domain PPG signal without the motion component, making it easier to identify the cardiac component of the PPG signal in the time domain. See Figure 2 below for a flowchart describing this filtering process.

Figure 2. Flowchart describing the process of isolating heart rate from PPG signal [1].

The processor does not need to identify frequencies of the motion signal for every time point. It identifies these frequencies within the PPG signal for an initial time period, configures a filter to remove these frequencies, then uses the same filter to filter the motion signal from subsequent time periods of the PPG signal.

The device is also capable of storing memory. This allows for the device to create a model of expected cardiac component frequencies from previously determined data. Based on the model, the processor can then determine the probability of any given frequency within the PPG signal to be a frequency of the cardiac component.

Heartbeat and respiratory patterns are cyclical over a short period of time while motion data and noise can be cyclical or irregular for any length of time. Over a longer period of time, cardiac and respiratory signals can potentially have non-cyclical patterns (e.g. increasing heart rate during an exercise session). This allows for the variability in cardiac parameters to be determined. Analyzing variability in heart rate allows for estimates of parameters of stress, recovery time, VO2 max, and sleep quality.


This patent cites numerous references of inventions this device incorporates or improves upon. This device improves on a previous wrist-watch heart rate monitor (patent 2009/0048526), which was developed as an alternative to wearing a chest strap heart rate monitor. The Garmin device is different from this wrist-watch as this device does not include any inertial sensors. Therefore the Garmin device is able to better remove noise from motion [2]. Another exercise device by Samsung Electronics (patent US 7,867,142 B2) uses heart rate data to inform users about changes in their exercise speed by playing a sound. While the Garmin device does not play a sound, it uses the heart rate data to extrapolate information about stress, recovery time, VO2max, and sleep quality, which is likely to be of greater value to the user [3].

The following lists basic information regarding the Garmin heart rate monitor patent:

  1. Patent title: Heart Rate Monitor With Time Varying Linear Filtering
  2. Patent number: US 9,801,587 B2
  3. Patent filing date: Oct. 18, 2016
  4. Patent issue date: Oct. 31, 2017
  5. How long it took for this patent to issue: 1 year, 13 days
  6. Inventors: Paul R. MacDonald, Christopher J. Kulach
  7. Assignee: Garmin Switzerland GmbH
  8. U.S. classification: CPC: A61B 5/02416 (20130101); A61B 5/1112 (20130101); A61B 5/1118 (20130101); A61B 5/7285 (20130101); A61B 5/721 (20130101); A61B 5/02405 (20130101); A61B 5/02427 (20130101); A61B 5/02438 (20130101); A61B 5/0833 (20130101); A61B 5/486 (20130101); A61B 5/4815 (20130101); A61B 5/681 (20130101); A61B 5/725 (20130101); A61B 5/7278 (20130101); A61B 5/165 (20130101); A61B 2562/0219 (20130101)
  9. How many claims: 29 claims



[1] P. R. MacDonald and C. J. Kulach, “Heart Rate Monitor With Time Varying Linear Filtering.” U.S. Patent 9,801,587 B2, issued October 31, 2017.

[2] R. M. Aarts and M. Ouwerkerk, “Apparatus for Monitoring A Person’s Heart Rate And/Or Heart Variation; Wrist-Watch Comprising The Same.” U.S. Patent 2009/0048526 A1, issued February 19, 2009.

[3] S. K. Kim, J. S. Hwang, and K. H. Kim, “Method and Apparatus for Managing Exercise State of User.” U.S. Patent 7.867,142 B2, issued January 11, 2011.

Patent Blog Post: Fitbit’s Wearable Heart Rate Monitor

Perhaps you’ve been barraged by emails from Fitbit that try and get you to buy one of their products during one of their many sales. Perhaps you’re a trendy techie and have a wearable in the form of a Galaxy or Apple Watch. Or perhaps you’re simply the owner of a smartphone made within the past few years. All these technologies have heart rate monitoring built into them from the get-go, and it is increasingly hard to get away from gadgets that don’t have some form of heart monitoring. With how ubiquitous the technology has gotten, I would like to look today at one of the patents put forward by Fitbit, one of the more popular brands when it comes to wearable fitness trackers. For this post, I’ll be using the information put forward by Google Patents, seen here.

One of the many figures in the patent, detailing the backside of the wearable.

The patent is simply titled as, “Wearable heart rate monitor,” and has a patent number of US8945017B2. It was originally filed on June 3rd, 2014, and was then approved on February 3rd, 2015. This makes the time to issue a little under a year, which is quite fast for an electronics product. The two inventors credited in the patent are Subramaniam Venkatraman and Shelten Gee Jao Yuen. Looking at the other patents associated with them, Venkatraman seems to have worked on more navigational devices, while Jao Yuen has worked on several other gyroscope-related projects. The assignee is, of course, Fitbit Inc. themselves. Officially, one of the classifications of the patent is, “signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal.” This one patent has 30 different claims to its name.

Of the 30 different claims in the patent, many of them tie into 2 main claims. The first is that the wearable heart monitor has a way to efficiently, accurately, and quickly determine the heart rate of the user. The second is to ensure that the wearable is capable of compiling the heart rate monitor’s data, including the heart rate data. This patent is aimed at both casual and advanced fitness enthusiasts, as the data gleaned from the wearable is handy to track. Runners, in particular, would find this tempting as it also mentions step tracking and other forms of movement.

The heart rate monitor works by using a waveform sensor, which reads signals at the surface of the skin. These signals are sent to the rest of the device, where the data is processed. The raw data from the sensor is rough and has a lot of noise from several factors, including movement and moisture. To remove the noise, the data has to be passed through several filters. From that data, a heart rate can be determined, and then presented to the user. Unlike the monitors of prior ages, this heart rate monitor would not rely upon disposable components, instead simply being able to be used multiple times by wearing it. In addition, the heart rate tracker would track more than just heart rate, including details about steps.


Venkatraman, S., & Yuen, S. G. J. (2014). Wearable heart rate monitor. Retrieved from

How It Works: Heart Rate Monitors

By: Margot Farnham, Juliana Gullotta & Nicholas Ruggiero

Suggested Reading: